The forgetting curve
In 1885 Hermann Ebbinghaus measured how much of a list of syllables he remembered after hours and days, and drew the first forgetting curve: a steep drop at first that then flattens out. More than a century of experiments has confirmed it with real material: what you learn today and never use again is mostly gone within a week.
That is exactly what happens with GeoGuessr metas. You read a bollard guide, understand all of it, and next game you can't tell whether the white bollard with a red band was Australia or Turkey. It's not a lack of attention: it's how memory works. It discards what it doesn't need again.
The coral curve is a video or a guide read once. The lime one is what happens when you review right before forgetting: every review resets the drop and, crucially, the next drop is slower. By the fourth or fifth review the clue survives months without being seen.
Spaced repetition
Spaced repetition is the study technique with the most evidence behind it: instead of reviewing everything every day, you review each thing right when you are about to forget it. It is the principle behind flashcards, Leitner boxes and programs like Anki or SuperMemo.
The advantage is twofold. First, efficiency: you waste no time on what you already know. Second, retention: a review done at the point of maximum difficulty (when it has almost slipped) fixes the memory far better than one done while it's still fresh. That's why MetaSpotter never shows you a card "just in case".
In practice: every card has a next-review date. Each day you see the ones that are due plus a handful of new ones (15 by default, adjustable). A typical session is ten minutes.
FSRS, the scheduler
The hard part is deciding when you are about to forget each clue. MetaSpotter uses FSRS (Free Spaced Repetition Scheduler), an open-source algorithm trained on hundreds of millions of real reviews, which Anki has included as an option since version 23.10. It is considerably more accurate than the classic 1980s SM-2 algorithm many apps still use.
FSRS models every card with three numbers:
- Stability: how many days until your chance of recalling it drops below 90%. It grows with every correct answer.
- Difficulty: how "sticky" that clue is for you. A bollard with two unique features is easy; telling Peruvian from Bolivian poles is hard.
- Retrievability: the probability that you remember it today, computed from the two above and the time since the last review.
The target is 90% retention: the app schedules each review for the day retrievability falls to that threshold. Answer correctly and stability grows, pushing the next review further away; miss it and the clue comes back later in the same session, then in short steps until it settles again.
The ladder above is illustrative but honest: 10 minutes → 1 day → 4 days → 12 days → 1 month → 3 months → 8 months is the real order of magnitude for a card you keep getting right. Seven reviews in total for a clue that no longer slips.
The four answers
When you get it right, the app shows the explanation and asks how well you knew it. That rating is all FSRS needs to decide when the clue comes back.
| Answer | When to use it | What the scheduler does |
|---|---|---|
| Guessed | Correct, but by luck | Counts as a miss: it comes back soon, in this same session too |
| Hard | Correct, but hesitant or very slow | A new clue comes back in 10 minutes, within the session if you keep going; one you were already reviewing, sooner than usual |
| Good | Correct with normal effort | A new clue comes back tomorrow; after that, the FSRS interval, longer every time |
| Easy | Instant, no thinking | It stays away for at least 30 days (you choose in Settings); in multiple choice, the clue moves on to a typed answer |
If you miss it, or press “I don’t know”, there is nothing to rate: you see the answer and the explanation, press Continue and the clue comes back a few clues later in the same session, so it sticks. If you used the hint (the light bulb), the rating is capped at Hard. On a computer you can rate with keys 1–4.
Answer modes
There are three card types, decided by the nature of each meta:
- Country: "In which country do you see this bollard?". Four choices (the distractors are the countries it is genuinely confused with, not four at random) or a typed answer with a corrector.
- Region: "Brazil: in which states do you see coffee fields?". You answer by tapping the country map; partial answers are accepted. This is the regionguessing mode.
- Identify: "Which camera generation is this?", "Which script is this?". Canonical text, multiple choice or typed.
Every card also carries an assist level that evolves with you: it starts with multiple choice and, once you get it right comfortably (two Goods in a row or one Easy), moves on to a typed answer with suggestions. A miss sends it back to the choices. If you prefer, Settings lets you fix one mode for every card: always choices or always typed, and for regions, map or list.
The typed-answer corrector understands accents, articles, abbreviations ("NSW") and names in English, Spanish and the local language, and asks "Did you mean Sulawesi?" if you drop a letter.
Learned, mastered and due clues
MetaSpotter’s progress keeps two things apart: what you have learned, which only grows, and what is due for review, which FSRS works out every day from each clue’s retrievability. That way you see how far you have come without losing sight of what is starting to slip. Every clue is in one of these states:
- New: you haven’t seen it yet.
- Learned: you have seen it at least once. It never goes back to new.
- Mastered: its next review is 21 days away or more, so it is solid for the long run. A miss sends it back to learned.
- Due: FSRS reckons today is the day to see it again so you don’t forget it.
That is what paints the map. Each country counts its country clues; region clues add up in their decks and in your collection, not in the country.
| Country | Definition | On the map |
|---|---|---|
| Unexplored | No clue learned yet | grey |
| Lit | At least one clue learned | lime, brighter the more you know |
| Explored | Every clue learned | lime |
| Mastered | Explored, with 90% of its clues mastered (in countries with 5 clues or more) | bright lime |
| Needs review | Reviews have piled up | amber border |
The Home and Progress maps show exactly that. Reviewing two hundred clues you already know inflates nothing: the map moves when you learn new clues and when you consolidate them, and the amber goes as soon as you do that country’s reviews.
Everything is retroactive and deterministic: it is computed from your review history, so if you import your progress on another device the map, rank and achievements appear instantly.
Review what’s slipping
When a country needs review, a line under the Home map says how many do and offers Review. It opens a session with only the clues that are due, from most forgotten to least, up to 50. On each country’s sheet, “Review” does the same for that country. It’s the most profitable session in the app: every correct answer there saves a clue that was about to go.
Streak and daily goal
The streak doesn't count days you open the app: it counts days you do your reviews. A day counts if nothing is left to review when you finish or, if a lot had piled up, once you have done 20 (change it in Settings, under “Reviews for your streak”). Every 7 consecutive days earns a streak shield, which is spent automatically on a day you miss; you can store one (three with Pro) and they are never for sale.
Ranks
There are no experience points. Rank comes from the clues you have learned and the countries you have explored and mastered, and it never drops: you keep the highest one you have reached, and what you forget is flagged for review. Each one describes something you can do in a game:
| Rank | Condition | What it means |
|---|---|---|
| Novice | — | You begin. |
| Explorer | 50 clues learned, 2 countries explored | You recognise the basics of a good handful of countries. |
| Cartographer | 300 clues learned, 10 countries explored, 2 mastered | You get the country in most rounds of Europe or Asia. |
| Tracker | 1,000 clues learned, 30 countries explored, 10 mastered | You tell bollards, poles and plates apart without thinking. |
| Master | 2,500 clues learned, 60 countries explored, 30 mastered | You play on par with people who have years of games. |
| MetaSpotter | 5,000 clues learned, 90 countries explored, 60 mastered, one region deck mastered | You know the meta before the round finishes loading: country and region. |
Achievements
Twelve achievements in four families. None rewards opening the app or doing N reviews: all of them measure a real skill or real consistency, and some are genuinely hard. Hidden ones are discovered.
First light
Continent lit
Around the world
Bollard eye
Pole reader
Plate spotter
Regionguesser
No longer confused
Nemesis
Steel memory
Map lit 7
Map lit 100Blue ring: easy. Lime: medium. Amber: hard. Violet: legendary. "Steel memory", for instance, demands 92% retention for 30 consecutive days; "Around the world", at least one clue alive on every continent.
FAQ
How much time a day?
About 10 minutes at the default pace of 15 new clues a day. If you have more time, raise your pace in Settings; on a busy day, just do your reviews.
What if I stop for a week?
Reviews pile up and some countries turn amber. You lose nothing: not what you learned, not your history, not your rank. When you come back, Review what’s slipping starts with what is most at risk, and within a few days you are back on track.
Is it the same as using Anki with a meta deck?
The scheduler can be the same if you turn on FSRS in Anki. What MetaSpotter adds is the content, ready made (14,516 metas with image, explanation and plausible choices), the interactive map for regionguessing, the typed-answer corrector and the whole per-country progress system.
Where do the metas come from?
From the open-source project Learnable Meta (AGPL licence), which in turn cites the Plonk It guides. Every card keeps its source. Images are Google Street View captures and carry their credit.